26419 research outputs found
Sort by
Aluminum as a Tracer of Dust Deposition to the Ocean: A Case Study from the Bermuda Region
No full textAluminum (Al), a major component of mineral aerosol (dust), partially dissolves in seawater and is widely used as a tracer for estimating time‐averaged dust fluxes to the ocean. Such estimates rely on dissolved Al (DAl) inventories in the surface mixed layer (SML), an assumed SML residence time of DAl (TDAl), the fractional solubility of Al in dust (AlS), and the mass fraction of Al in dust. In this study, dust flux estimated from seasonal, water-column DAl data from the Bermuda Atlantic Time-series Study (BATS) region are compared with direct dust flux estimated from contemporaneous measurements of Al in aerosols and rain collected at Tudor Hill, Bermuda, over a 318-day period. The DAl-based flux estimates, using a 200 m deep SML, range from 5.3–11.2 g m⁻² yr⁻¹, which is substantially higher than the flux estimate of 1.2 g m⁻² yr⁻¹ based on Al in aerosols and rain. This discrepancy likely results from an underestimate of ʈDAI, as well as the influence of lateral transport over the longer timescale of the true τDAl value. A seasonally constrained DAl-based flux estimate, limited to the stratified summer months (April–August), yielded fluxes of 1.5–6.3 g m⁻² yr⁻¹, which are more consistent with the directly measured summer-period deposition of 1.31 g m⁻² yr⁻¹. These results highlight the importance of seasonal dynamics, temporal context and physical transport in tracer-based estimates of dust deposition in the open ocean
Review of \u3ci\u3eInternet Invention: From Literacy to Electracy\u3c/i\u3e, by Gregory L. Ulmer, 2003
No full textS²IL: Structurally Stable Incremental Learning
No full textFeature Distillation (FD) strategies are proven to be effective in mitigating Catastrophic Forgetting (CF) seen in Class Incremental Learning (CIL). However, current FD approaches enforce strict alignment of feature magnitudes and directions across incremental steps, limiting the model’s ability to adapt to new knowledge. In this paper, we propose Structurally Stable Incremental Learning (S²IL), a FD method for CIL that mitigates forgetting by focusing on preserving the overall spatial patterns of features which promote flexible (plasticity) yet stable representations that preserve old knowledge (stability). We also demonstrate that our proposed method S²IL achieves strong incremental accuracy and outperforms other FD methods on SOTA benchmark datasets CIFAR-100, ImageNet-100 and ImageNet-1K. Notably, S²IL outperforms other methods by a significant margin in scenarios that have a large number of incremental tasks. The source code is available at
https://github.com/dlclub2311/Structurally-Stable-Incremental-Learning
AI-Based Steganography Method to Enhance the Information Security of Hidden Messages in Digital Images
No full textWith the increasing sophistication of Artificial Intelligence (AI), traditional digital steganography methods face a growing risk of being detected and compromised. Adversarial attacks, in particular, pose a significant threat to the security and robustness of hidden information. To address these challenges, this paper proposes a novel AI-based steganography framework designed to enhance the security of concealed messages within digital images. Our approach introduces a multi-stage embedding process that utilizes a sequence of encoder models, including a base encoder, a residual encoder, and a dense encoder, to create a more complex and secure hiding environment. To further improve robustness, we integrate Wavelet Transforms with various deep learning architectures, namely Convolutional Neural Networks (CNNs), Bayesian Neural Networks (BNNs), and Graph Convolutional Networks (GCNs). We conducted a comprehensive set of experiments on the FashionMNIST and MNIST datasets to evaluate our framework’s performance against several adversarial attacks. The results demonstrate that our multi-stage approach significantly enhances resilience. Notably, while CNN architectures provide the highest baseline accuracy, BNNs exhibit superior intrinsic robustness against gradient-based attacks. For instance, under the Fast Gradient Sign Method (FGSM) attack on the MNIST dataset, our BNN-based models maintained an accuracy of over 98%, whereas the performance of comparable CNN models dropped sharply to between 10% and 18%. This research provides a robust and effective method for developing next-generation secure steganography systems
Protonic Capacitor Cell Energetics: Transmembrane-Electrostatically Localized Protons/Cations
No full textThe transmembrane-electrostatically localized protons/cations charges (TELPs/TELCs) theory can serve as a theoretical framework to better explain cell electrophysiology and elucidate bioenergetic systems including both delocalized and localized protonic coupling. According to the TELCs model, the excess positive charges of TELCs at one side of the membrane are balanced by the excess negative charges of transmembrane-electrostatically localized hydroxides anions (TELAs) at the other side of the membrane. Through the TELCs-membrane-TELAs capacitor model, the energetics of oxidative phosphorylation have recently been better elucidated in mitochondria and alkalophilic bacteria, leading to the identification of a novel Type-B energetic process. Both the TELCs model studies and experimental demonstration results showed that the putative “potential well/barrier” model is not needed to explain TELPs formation. Application of the TELCs model to neural cells has recently resulted in novel neural transmembrane potential integral equations. In this review article, we will visit the TELCs-membrane-TELAs model and its applications including its features and predictions that may help better understand cell energetics. Meanwhile, we will also discuss some of the recent critiques and point out the opportunities and directions for future research. The TELCs model can be well predictive and provide new opportunities as a theoretical tool for further research to better understand cell physiology, bioenergetics, and neurosciences. This Landmark Review article timely provides the latest discoveries, breakthrough advances with new developments and knowledge, directions and opportunities for future research in a major emerging and exciting scientific area of protonic capacitor cell energetics: transmembrane-electrostatically localized protons/cations
Doctoral Student Supervisors\u27 Experiences with Broaching Race and Race-Related Issues
No full textIn this descriptive phenomenological study, we explored 13 doctoral student supervisors’ experiences with broaching race and race-related issues in supervision with counselors-in-training. Three themes emerged: the function of broaching; self-perceived attributes affecting broaching; and experiences of programmatic support and challenges. We discussed practical and research implications with study limitations
Influence of Gender-Specific Data Imbalance on scGPT Fine-Tuning for Single-Cell Genomics
No full textThe transformer-based foundation model scGPT has demonstrated strong capabilities in analyzing high-dimensional single-cell RNA sequencing data. However, the impact of demographic factors, particularly gender, on model performance remains insufficiently understood. Gender is known to influence cell type compositions in the immune system. Here, using the gender-sensitive cell type composition in immune system, we comprehensively evaluate how the gender-sensitive imbalance of training data influences the performance of scGPT in cell type predictions. We fine-tune scGPT on male-only, female-only, and mixed-gender subsets from two large-scale datasets containing immune cells. We use a logit difference to measure the confidence gap between the true label and the actual model prediction. The confidence gap is zero for perfect classifications and negative for incorrect predictions. We find that training and testing configurations with aligned gender distributions generally show higher prediction confidence, while mismatched gender during training and testing, especially when training excludes one gender, leads to substantial confidence drops. We also find that training with mixed-gender data promotes more balanced generalization, but does not eliminate all biases. We conclude that gender-specific data imbalance, represented by immune cell type subpopulation variation between women and men, can influence fine-tuning of scGPT and its performance in cell type classification, highlighting the importance of addressing such demographic biases in biomedical AI models
Shard-Unlearn: A Sharded Elastic SGD Privacy Preserving Federated Unlearning Framework for 5G-Assisted Healthcare
No full textSmart healthcare systems are generating unprecedented volumes of sensitive data, making robust privacy preservation a critical requirement. Traditional machine unlearning (MU) techniques aims to excise specific data points and their statistical influence from trained machine learning (ML) model. Thus, they suffer from limited computational efficiency, poor scalability, and suboptimal model convergence when applied to largescale, big-data (BD) healthcare environments. These limitations become even more significant in 5G-assisted settings, where real-time connectivity and rapid data processing are essential. To address these challenges, we introduce the concept of data sharding which partitions healthcare datasets into manageable segments. In the paper, we introduce Shard-Unlearn framework, that implements federated unlearning (FU) process to the shards that contain sensitive data. This reduces the overall computational overhead and optimizes model convergence over 5G networks. In the framework, we present the elastic stochastic gradient descent (SGD) optimization which effectively remove the targeted data and associated statistical perturbations from the local models. The framework is tested over the ADMISSIONS benchmark dataset, which is divided 10 shards. The framework is compared on computational efficiency, model robustness, and privacy preservation metrics. Statistical findings reveals a 47.14% improvement in unlearning impact (as measured by recall) while striking a balanced trade-off between performance and data security. These results underscore the viability of the framework as a scalable and privacy-preserving solution formodern 5G-assisted healthcare systems
A Global Application Programming Interface-Enabled Earthquake Ground Motion Relational Database for Engineering Applications
No full textWe present a application programming interface (API)-enabled relational database of global earthquake ground motion intensity measures, associated metadata, and processed time-series data. Raw ground motion records were processed by the authors using either manual or semi-automated processing procedures, and every processed record has passed a quality review by a trained analyst. Computed intensity measures include peak acceleration and velocity, pseudo-spectral acceleration response spectra, cumulative absolute velocity, Arias Intensity, and Fourier amplitude spectra. The processed time-series data, associated metadata, and ground motion intensity measures were organized into a web-served relational database consisting of 32 tables connected by primary/foreign key pairs. Ground motion metadata and intensity measures (but not time-series) from the Next-Generation Attenuation (NGA)-East and NGA-West2 projects and the Hellenic Strong-Motion Database are also contained in the database. As of this writing (June 2025) the database includes intensity measures and metadata for 76,242 multi-component ground motions recorded at 9927 stations for 1391 events, and is approximately 73.5 GB in size. The database is built using the MySQL relational database management system, and is accessible through a web interface and also an API, which allows users to retrieve data using straightforward and intuitive uniform resource locators (URLs). Compared with more traditional file-download-based methods for data release, the relational database (1) increases storage efficiency, (2) improves data integrity, and (3) enables users to query the data subset they wish to retrieve rather than downloading the entire database and loading it into memory. Furthermore, the web-served nature of the database means that users have immediate access to ground motion data following collection, review, and uploading. Periodic static releases of the database will be published as a means of archiving and facilitating reproducibility. The database has been designed to accommodate growth, with ongoing efforts to integrate global ground motion data (e.g. data development for the NGA-West3 project)
Building Cyber Resilience: Educational Programs in K-12 Education
No full textThe article discusses the challenges of teaching cybersecurity in K-12 education. Topics mentioned include the lack of access to resources and appropriate professional development, the career and technical education cybersecurity pathways, the fundamental pathways for cybersecurity and information technology and the results of program evaluation in several local community high schools from 2021 and 2022